Design and Implementation of Model-Based and Learning Control for Rehabilitation Parallel Robots:: Advancements and Performance Evaluation
- Autores: Rafael J. Escarabajal
- Directores de la Tesis: Marina Vallés (dir. tes.), Ángel Valera Fernández (dir. tes.)
- Lectura: En la Universitat Politècnica de València ( España ) en 2024
- Idioma: español
- Tribunal Calificador de la Tesis: Erika Ottaviano (presid.), Antonio José Sánchez Salmerón (secret.), Concepción Alicia Monje Micharet (voc.)
- Programa de doctorado: Programa de Doctorado en Automática, Robótica e Informática Industrial por la Universitat Politècnica de València
- Materias:
- Enlaces
- Tesis en acceso abierto en: RiuNet
- Resumen
The current era is characterized by a growing elderly population due to increased life expectancy and a declining birth rate. This demographic shift leads to a rise in disabilities, emphasizing the importance of healthy aging and functional ability. In response to these challenges, innovative assistive methodologies are being explored, particularly in the integration into physical human-robot interaction.
Parallel Robots (PRs) are closed kinematic chains that exhibit unique advantages for human limb rehabilitation, thanks to their stiffness, accuracy, and robustness. However, their complex modeling and the presence of singularities within the workspace pose a challenge to ensure safe and effective interaction. This document proposes a collection of novel control algorithms using a PR to cover a wide spectrum of applications within the context of assistive robotics, ranging from passive training (in which the user does not generate the movements, so minimal to no effort is induced) to active exercises (requiring the patient's voluntary muscle contractions) and power augmentation.
The first algorithm addresses the challenge of trajectory tracking with a model-based controller with force exchange between the human and the robot in the absence of a force sensor. By including a mechanism to estimate this interaction force online using Least Squares (LS) within the model, the algorithm provides valuable insight for therapists or other algorithms and continuously reduces tracking errors by correcting the dynamic model online.
In active training scenarios, force-based controllers are essential to respond appropriately to interactions. The thesis incorporates an admittance controller with a force sensor. Despite the benefits of this interaction, there are risks associated with human modification of robot trajectories due to the presence of singularities within the workspace that must be avoided to ensure the patient's safety. Dynamic Movement Primitives (DMP) are employed to encode the initial trajectory, which can be modified with coupling terms to achieve both objectives of admittance control and singularity avoidance simultaneously.
In the early stages of rehabilitation, patients may lack full motor skills due to the injury, so passive exercises should be employed. This research proposes an intelligent mechanism for passive exercises that allows users to return to previous safe positions and resume the exercise in a self-paced manner. The trajectory reversal is achieved with Reversible Dynamic Movement Primitives (RDMP). The approach involves encoding the expected behavior using data from the analogous healthy limb and reversing the trajectory when the injured limb deviates from the established statistical patterns. This approach aligns with the paradigm of Imitation Learning.
Finally, musculoskeletal models play a crucial role in estimating the user's muscle forces, offering significant utility when integrated in controllers operating in muscular space. These controllers can be applied in both power augmentation and rehabilitation contexts. In power augmentation, the concept of manipulability describes the human's ability to exert forces in any direction. Achieving isotropic manipulability is desirable to maintain constant muscular activation, and this study investigates its representation with a novel concept called a force envelope. Furthermore, this research explores the use of muscle-targeted controllers in assistance or rehabilitation settings, by means of a closed-loop controller designed to induce specific tension forces in muscles.
